Deep Learning - Internship

Machine Learning Master (last year). Autonomous, creative, interested in academia.

JOB DESCRIPTION

Deep Learning Internship : Segmentation and localization at Heuritech

Context

Heuritech builds and commercializes a software that understands unstructured data, such as text and images. It uses state of the art Deep Learning expertise. Our main application is trend detection on the web, including Fashion trends seen on images.

Mission

Collaborating with engineers and scientists in order to develop new services and improve current ones, in fields such as classification, localization and segmentation of image data. Focus on clothing, fashion and beauty image analysis.

Creating and implementing cutting edge deep learning and computer vision algorithms. Comparing different algorithms such as (but not restricted to) fully convolutionnal networks, Faster R-CNN, SegNet, CRF based algorithms, mixed models.

Experimenting new ideas and participate to the creative processes.

Profile

Machine Learning Master (last year). Autonomous, creative, interested in academia.

  • Good knowledge of Python, Linear algebra
  • Knowledge of at least one of Theano/Torch/TensorFlow.

Internship information

  • Starting date: September-October 2016

  • Length: 4-6 months

  • Place: Heuritech, 248 rue du faubourg saint Antoine, Paris

  • Compensation: Depends on profile

References

Noh, Hyeonwoo, Seunghoon Hong, and Bohyung Han. "Learning deconvolution network for semantic segmentation." Proceedings of the IEEE International Conference on Computer Vision. 2015.

Liu, Si, et al. "Fashion parsing with weak color-category labels." IEEE Transactions on Multimedia 16.1 (2014): 253-265.

Simo-Serra, Edgar, et al. "A high performance CRF model for clothes parsing." Asian conference on computer vision. Springer International Publishing, 2014.

Team

The intern will join the Heuritech team, a company focused on Machine Learning. The team consists of 15 employees, including 7 PhDs in Machine Learning and AI. 

Please send your resume and motivations to contact@heuritech.com.

Machine Learning - Internship

Machine Learning Master (last year). Autonomous, creative, interest in academia.
Knowledge of Python, Linear algebra, Theano and/or Torch.

JOB DESCRIPTION

Machine Learning Internship : Deep Learning on textual data at Heuritech

Context

Textual data is inherently sequential, and several Machine Learning advances have been found to process it (such as Markov Processes, Deep Recurrent Networks, ...).
When building a classification model on short and complex text such as URLs, the extraction of words can be difficult. The use of character level classification can increase performance and generalization to unseen data. The goal of this internship is to explore character level classification and compare it to more standard approaches.

Mission
Collaborating with engineers and scientists in order to develop new services and improve current ones, in fields such as classification and exploration of textual data. 
Creating and implementing cutting edge deep learning algorithms for classification of textual data. Comparing different algorithms such as (but not restricted to) bag of ngrams, recurrent neural networks (GRU, LSTM), 1d convolutional neural networks.
Experimenting new ideas and participate to the creative processes.

Profile
Machine Learning Master (last year). Autonomous, creative, interested in academia. Knowledge of Python, Linear algebra, Theano and/or Torch.

References

Text Understanding from Scratch, Xiang Zhang, Yann LeCun, 2015
Character-level Convolutional Networks for Text Classification, Xiang Zhang, Junbo Zhao, Yann LeCun, 2015
Example use of RNN on character level : http://karpathy.github.io/2015/05/21/rnn-effectiveness/

Team
The intern will join the Heuritech team, a company focused on Machine Learning. The team consists of 10 employees, including 6 PhDs in Machine Learning and AI. 

Heuritech builds and commercializes a software that understands unstructured data, such as text and images. The software uses advances Deep Learning techniques and is already in production among several customers. Our goal is to match a human level of understanding on any textual/image data, across different domains and languages.